卷积神经网络在结构损伤诊断中的应用

对卷积神经网络(CNN)在工程结构损伤诊断中的应用进行了深入探讨; 以多层框架结构节点损伤位置的识别问题为研究对象,构建了可以直接从结构动力反应信号中进行学习并完成分类诊断的基于原始信号和傅里叶频域信息的一维卷积神经网络模型和基于小波变换数据的二维卷积神经网络模型; 从输入数据样本类别、训练时间、预测准确率、浅层与深层卷积神经网络以及不同损伤程度的影响等多方面进行了研究。结果表明:卷积神经网络能从结构动力反应信息中有效提取结构的损伤特征,且具有很高的识别精度; 相比直接用加速度反应样本,使用傅里叶变换后的频域数据作为训练样本能使CNN的收敛速度更快、更稳定,并且深层CNN的性能要好于浅层CNN; 将卷积神经网络用于工程结构损伤诊断具有可行性,特别是在大数据处理和解决复杂问题能力方面与其他传统诊断方法相比有很大优势,应用前景广阔。     

基于人工神经网络的土工合成材料加筋挡墙临界高度预测模型(英文)

提出了一种基于人工神经网络(ANN)技术的加筋挡墙设计高度预测方法。通过分析挡墙失效的原因,确定了7个主要因素作为网络的输入神经元。收集23组挡墙离心模型试验数据,2组足尺试验数据,1组实际工程的破坏数据,共26组样本作为训练及检验样本,建立了可用于加筋挡墙设计高度预测的径向基函数网络(RBFN)及误差反传网络(BPN)模型。结果表明径向基函数网络在学习速度,预测准确性及网络推广能力方面均优于BP网络,本文方法可用于加筋支挡结构的设计参考。     

大跨度空间网格结构的损伤定位

本文建立了基于模态曲率法和人工神经网络技术相结合的、适用于大跨度空间网格结构的损伤定位新方法,即首先应用模态曲率法判断结构是否发生损伤并识别发生损伤的局部结构,然后对发生损伤的局部结构利用人工神经网络技术识别损伤的准确位置。通过分析和比较发现,以模态曲率为基础的损伤参数比较适合于大跨度空间网格结构的损伤定位,三种以模态曲率为基础的损伤定位参数按有效性进行排序,从低到高依次为模态曲率、模态曲率差、模态曲率变化率;针对天津奥林匹克中心体育场大跨度悬挑管桁结构进行了不同损伤状况的数值模拟,验证了所建立的损伤定位方法的适用性和有效性。研究结果表明:利用模态曲率变化率识别损伤发生的大致位置,当单榀桁架发生损伤时,识别的准确率达到100%,当多榀桁架同时发生损伤时,识别的准确率达93.7%;采用人工神经网络技术识别损伤桁架的准确损伤位置时,在无测量噪声影响下,损伤定位的准确率达到97.0%,且测量噪声对损伤定位准确率的影响很大。     

基于模态刚度的预应力混凝土梁损伤识别方法研究

阐明了模态刚度在损伤识别研究中的重要意义,并对11根多级损伤状态的预应力混凝土梁进行动力试验研究。通过对梁模态分析发现,由于噪音污染等多种因素的影响,仅凭各梁实测模态刚度数值的直观分析很难对梁的多级损伤状态进行有效的识别。为此,提出了以模态刚度变化率为损伤指标的BP神经网络和PNN神经网络的损伤识别方法,并利用实测数据验证所提方法的实用性。研究表明,两种神经网络分类器识别方法均能够有效应用于实际中,且具有很高的损伤识别精度,为结构损伤识别方法研究提供了新思路。     

Patterning and predicting aquatic macroinvertebrate diversities using artificial neural network

A counterpropagation neural network (CPN) was applied to predict species richness (SR) and Shannon diversity index (SH) of benthic macroinvertebrate communities using 34 environmental variables. The data were collected at 664 sites at 23 different water types such as springs, streams, rivers, canals, ditches, lakes, and pools in The Netherlands. By training the CPN, the sampling sites were classified into five groups and the classification was mainly related to pollution status and habitat type of the sampling sites. By visualizing environmental variables and diversity indices on the map of the trained model, the relationships between variables were evaluated. The trained CPN serves as a 'look-up table' for finding the corresponding values between environmental variables and community indices. The output of the model fitted SH and SR well showing a high accuracy of the prediction (r>0.90 and 0.67 for learning and testing process, respectively) for both SH and SR. Finally, the results of this study, which uses the capability of the CPN for patterning and predicting ecological data, suggest that the CPN can be effectively used as a tool for assessing ecological status and predicting water quality of target ecosystems.     

A wavelet-based technique for damage quantification via mode shape decomposition

In this study, a neuro-wavelet technique was proposed for damage identification of cantilever structure. At first, damage localisation was accomplished through mode shape decomposition using discrete wavelet transforms. Subsequently, a damage indicator was defined based on the detail coefficients of the decomposed signals. It was found that distinct patterns relate the damage indicators to damage locations. Considering this property, a neural network ensemble was developed for damage quantification. Damage indicators and damage locations were selected as input parameters for the neural networks. Three individual neural networks were trained by input samples obtained from different combinations of decomposed mode shapes. Then, the outcomes of the individual neural networks were fed to the ensemble neural network for damage quantification. The proposed method was tested on a cantilever structure both experimentally and numerically. Six different damage scenarios including three different damage locations and three different damage severities were introduced to the structure. The results revealed that the proposed method was able to quantify different damage levels with a good precision.     

基于模态应变能与神经网络的钢网架损伤检测方法

神经网络通过对样本的学习,获得结构模态参数与损伤之间的映射关系。目前基于神经网络的损伤检测已经越来越广泛地使用在非破坏性损伤诊断当中。但对于大型结构而言,它的训练样本数量过大,将消耗大量的计算。所以如何降低神经网络的计算量使其可用于大型结构的损伤诊断是一个亟待解决的问题。为了解决这个问题,提出了空间钢网架损伤的两步诊断法:第一步,利用模态应变能对结构损伤的敏感性,判断出结构损伤的可能位置;第二步,利用神经网络从可能发生损伤的杆件中定位出实际损伤的位置,并进行损伤程度的判断。利用一个空间网架作为数值算例,进行可行性验证。结果表明此方法可以准确判断出结构的损伤位置以及损伤大小,是一种行之有效的方法。     

Hybrid intelligent method optimization of a soft rock replacement scheme for a large cavern excavated in alternate hard and soft rock strata

Optimization of a soft rock replacement scheme for a large cavern excavated in alternating hard and soft rock strata is a complicated non-linear mechanical problem having a large parameter search space. Obtaining a global optimum solution is the key to the problem. A hybrid intelligent method is proposed for this purpose. It is an integration of an evolutionary neural network and finite element analysis using a genetic algorithm. The non-linear relation of the soft rock replacement scheme with the displacement and damage zone of the cavern due to excavation in the given geological conditions is learnt and represented by a forward neural network whose structure and connection weights are global optimally recognized by using the genetic algorithm. The learning samples are obtained from finite element calculations. The optimal soft rock replacement scheme, having the minimal displacement and damage volume induced by cavern excavation, is searched in a global space using the genetic algorithm. The new methodology is used to evaluate soft rock replacement schemes for the Shuibuya cavern in China excavated in strata consisting of alternating soft and hard rocks. The results indicate that the new methodology can recognize the optimal soft rock replacement scheme for a large cavern in such complicated geological conditions and the neural network model can provide a solution which is close to the finite element analysis for the same geological and construction conditions.     

Dynamic Wavelet Neural Network for Nonlinear Identification of Highrise Buildings

Abstract:   Recently, the authors presented a multiparadigm dynamic time-delay fuzzy wavelet neural network (WNN) model for nonparametric identification of structures using the nonlinear autoregressive moving average with exogenous inputs. Compared with conventional neural networks, training of a dynamic neural network for system identification of large-scale structures is substantially more complicated and time consuming because both input and output of the network are not single valued but involve thousands of time steps. In this article, an adaptive Levenberg–Marquardt least-squares algorithm with a backtracking inexact linear search scheme is presented for training of the dynamic fuzzy WNN model. The approach avoids the second-order differentiation required in the Gauss–Newton algorithm and overcomes the numerical instabilities encountered in the steepest descent algorithm with improved learning convergence rate and high computational efficiency. The model is applied to two highrise moment-resisting building structures, taking into account their geometric nonlinearities. Validation results demonstrate that the new methodology provides an efficient and accurate tool for nonlinear system identification of high-rising buildings.     

用于结构损伤识别的神经网络输入选取规则探究

将神经网络作为模式识别工具用于结构损伤位置识别时,其识别效果除了要受到网络隐层数目、各隐层神经元数目、神经元传递函数的形式、训练样本的数量与质量及训练方法的影响外,还会受网络输入性能的影响。在其他因素均相同的条件下,网络输入对网络性能起着决定性作用。为解决网络输入的选取问题,从网络功能、类别可分性和噪声的影响三个方面对网络输入的选取进行了分析研究,提出了用于结构损伤识别的神经网络输入选取的一般性规则,对采用神经网络处理模式识别问题具有参考价值。     

神经网络在结构工程健康监测中的应用探讨

建立结构损伤诊断子系统是建立大型工程结构智能健康监测专家系统的核心问题。人工神经网络技术可以实现结构损伤的自动识别与定位,具有广阔的应用前景。本文介绍基于人工神经网络的两级损伤识别策略,并对采用人工神经网络进行结构损伤诊断的网络输入参数与网络结构选择等关键问题进行了探讨。     

基于区间估计的结构损伤识别方法研究

当检测样本与训练样本的噪声水平不同时,用“自联想”神经网络对结构损伤存在性进行识别时会出现正误判,为解决这一问题,提出用区间估计的方法对结构损伤存在性进行识别,给出了2种情况下区间估计的计算方法。研究表明：当结构没有损伤时,即使检测状态的噪声水平与初始完好状态的噪声水平不同,区间估计的方法也能给出较高正确率的识别结果;当损伤达到一定程度时,只需要相对较少的样本,即可由区间估计的方法对结构是否发生损伤做出正确的识别。     

桁架结构损伤识别的数据融合方法研究

以数据融合技术进行桁架结构的单损伤和多损伤识别。通过研究基于频率的结构损伤理论,分析归一化的频率和损伤位置的关系;利用小波概率神经网络的算法对决策融合进行修正,建立基于小波概率神经网络的数据融合结构损伤识别模型。运用结构计算软件计算了一典型桁架结构的频率,并融合为小波概率神经网络算法的输入特征向量,并对桁架算例模型结构进行损伤识别。通过桁架不同位置的损伤情况,验证该方法的有效性,并提出工程应用中应注意的问题。研究结果表明,基于小波概率神经网络算法的数据融合技术是一种比较可靠的损伤识别方法,具有良好的工程应用前景。     

基于神经网络的结构损伤识别研究进展

为提高结构损伤识别方法的精确性和适用性,将神经网络引入到结构损伤识别中。介绍了神经网络的由来、原理和研究意义,概述了国内外基于神经网络的结构损伤识别研究进展。通过分析可以看,出用于结构损伤识别的神经网络方法有着广阔的应用前景。论文针对进一步研究的方向提出了建议。     

索膜结构风振响应的神经网络辅助参数分析方法

针对非线性动力时程分析法求解大规模索膜结构风振响应时动力时程分析的次数受到限制而导致一些参数组合下的响应统计值难以预测的问题,引入神经网络,通过少量样本的训练,建立了参数与结构响应间的映射关系。结果表明:提出的神经网络辅助参数分析方法计算效率高、预测精度令人满意,是一种获取足够数据的有效途径;通过该方法可以得到响应统计量及风振系数随平均风速和索、膜预应力变化的规律,为设计风荷载和结构构件极端响应的计算提供了科学依据。     

模型参数误差对用神经网络进行结构损伤识别的影响

通过理论推导得到了模型参数误差对损伤引起模态参数改变的贡献的表达式,用该式可指导神经网络输入参数的选择和输入向量的构造.理论分析表明,适当地构造输入向量,可以减小模型参数误差对结构损伤识别的影响.在采用BP网络和合适的输入向量后,还用数值模拟的方式对一榀六层框架的损伤识别进行了确定性研究和概率分析,结果表明,用神经网络进行结构损伤识别,受模型参数误差的影响很小,在训练神经网络时,10%的模型参数误差是可以接受的.最后,用一个两层钢框架的实验数据验证了神经网络在有模型误差时的识别能力.     

基于均匀设计实验的框架结构损伤识别应用研究

为对不规则框架结构的柱进行损伤识别,由位移指标对损伤初步定位后,利用均匀设计思想构造分步识别方案,通过BP神经网络利用频率和位移指标对损伤构件进行具体定位。通过6层框架数值模拟结果表明此方法克服了大型均匀设计表偏差过大,以及单一参数进行损伤识别的缺陷,能够较好地对框架结构损伤位置进行精确判定。     

基于前馈网络的岩体爆破效应预测研究

将神经网络理论知识和爆破专业知识有机地结合在一起，提出了一种新的岩体爆破效应预测的前馈网络理论方法。该方法适合于不同的爆破参数和不同的岩体条件，是一种普遍适用的方法，同时也是一种“面向数据”的方法。通过对三峡工程左岸坝区岩体爆破效应预测的研究表明，本文方法与通常的经验公式法、回归分析法以及ＢＰ网络方法相比，具有较高的预报精度     

Exploring the possibility of assessing the damage degree of liquefaction based only on seismic records by artificial neural networks

This study presents a new approach to determine the damage degree of liquefaction caused by a large earthquake. We propose an artificial neural network (ANN) model based only on the seismic records of ground and define the degree of liquefaction “DDL” as a damage index. This ANN model predicts the degree of excess pore water pressure increase as the correct output label based on the seismic records obtained from the three-dimensional shaking table test. The proposed model achieved high accuracy, and the outcomes from training data indicated that the ANN model is suitable to function as a liquefaction assessment system. Further, to evaluate the applicability of the proposed ANN model in the real world, the datasets of waves from three actual seismic records were input to the ANN as validation data. The DDL judgment obtained was a good fit with the real phenomena observed.     

Hybrid CPN–Neural Dynamics Model for Discrete Optimization of Steel Structures

Abstract: In practical design of steel structures, the designer usually must choose from a limited number of commercially available shapes such as the widely used wide flange shapes. In this article, we present a hybrid counterpropagation-neural dynamics model and a new neural network topology for discrete optimization of large structures subjected to the AISC ASD specifications. The constrained structural optimization problem is formulated in terms of a neural dynamics model with constraint and variable layers. The counterpropagation part of the model consists of the competition and interpolation layers. The CPN network is trained to learn the relationship between the cross-sectional area and the radius of gyration of the available sections. The robustness of the hybrid computational model is demonstrated by application to three examples representing the exterior envelope of high-rise and super-high-rise steel building structures, including a 147-story structure with 8904 members.